From fragments to digital wholeness: An AI generative approach to reconstructing archaeological vessels

Abstract

Reconstructing archaeological vessels from their fragments is a complex task that requires a long investment of time as well as in-depth knowledge of specific archaeological material. This paper proposes a framework based on generative artificial intelligence to reconstruct the entire vessel from a fragment. The proposed framework is based on a fragment simulation mechanism and the combination of three different deep learning models that position, reconstruct, and post-process the fragment to obtain a ready-to-use reconstruction. The method is applied as a case-study to a dataset of six Italian Bronze and Early Iron Age burial contexts, including about 4000 complete vessels and over 400 actual fragments. The results are evaluated using statistical metrics and expert human evaluation, showing promising results. The proposed method is a positive application of generative artificial intelligence in archaeology and provides a solution to the use of fragments in the digital and computational analysis of ceramics. The dataset, as well as the code used and the analytical pipeline, are fully available in the supplementary materials.